AI in Schools: What Educators Are Learning

When ChatGPT launched in late 2022, many schools responded with alarm, and some school systems blocked access on their networks. Several of those bans were reversed within months. New York City, the largest US school district, restricted ChatGPT in January 2023 and reversed course in May, when its chancellor wrote that the initial caution had evolved into "an exploration and careful examination" of the technology. Since then, the conversation in education has shifted from whether students will use AI to how they should, and what schools need to change as a result.

Updated September 2026: we added New York City’s ChatGPT ban and reversal, UNESCO’s guidance, an example of a study mode and research on detector bias.

How teachers are using AI

Much of the practical adoption has been by teachers themselves:

• Planning: drafting lesson plans, activities and differentiated materials for different reading levels
• Assessment: generating quiz questions and rubrics, and giving first-pass feedback on drafts
• Administration: writing emails, newsletters and reports
• Accessibility: simplifying texts, translating materials for families and supporting students with additional needs

Teachers who use these tools often describe time savings on preparation and paperwork. As with any AI output, materials still need checking for accuracy and suitability.

How students are using it

Students use AI for explanations, brainstorming, feedback, studying and, sometimes, doing the work for them. The same tool can be a tutor or a shortcut, depending on how it is used. Several AI companies have introduced study or learning modes designed to guide students through problems with questions rather than handing over answers, such as ChatGPT’s study mode, launched in July 2025.

The shift from bans to guidance

Many education authorities and organizations, including UNESCO, have published guidance encouraging a human-centered approach: clear rules, teacher training, age-appropriate use and data protection, rather than blanket prohibition. UNESCO’s 2023 guidance, for example, recommends a minimum age of 13 for using AI tools in the classroom, and noted that fewer than 10% of schools and universities had formal policies on generative AI. A common model sets AI rules per assignment, ranging from “no AI” to “AI allowed with disclosure”.

What is working

• Clear, specific expectations for each task
• Assessing process, including drafts, reflections and oral explanations
• Teaching AI literacy: how these tools work, how they fail and how to verify them. Our hallucinations guide is one example of what students need to know.
• Redesigning assignments to require personal perspective, local context or in-class work

What is not working

• Relying on AI detectors, which produce false positives and can unfairly penalize students (a Stanford study found that seven detectors flagged 61% of essays by non-native English speakers as AI-written), as explained in our detectors guide
• Policies that differ unpredictably between classrooms without explanation
• Ignoring the issue, which leaves students guessing about what is allowed

Open questions

Other unresolved issues include equity, since students with paid tools and supportive guidance may benefit most; privacy, particularly for younger children; and teacher workload, since redesigning assessments takes time schools rarely have.

Neither bans nor blind adoption

Schools are learning that AI cannot be banned away, and that it should not be adopted uncritically either. The most promising approaches treat AI like any powerful tool: taught explicitly, used deliberately and never allowed to replace the thinking students are there to develop.

Sources

1. ChatGPT caught NYC schools off guard. Now, we’re determined to embrace its potential., Chalkbeat, May 2023
2. UNESCO: Governments must quickly regulate generative AI in schools, UNESCO, September 2023
3. Introducing study mode, OpenAI, July 2025
4. AI detectors biased against non-native English writers, Stanford HAI, 2023